Modeling Individual Brain Networks and Nonlinear Dynamics with Diffusion Models
Poster No:
1467
Submission Type:
Abstract Submission
Authors:
Meenu Ajith1, Vince Calhoun2
Institutions:
1Georgia State University, Atlanta, GA, 2GSU/GATech/Emory, Atlanta, GA
First Author:
Co-Author:
Introduction:
Functional brain networks exhibit distinct spatial and temporal patterns varying within and across individuals. Precise modeling of individualized brain network models can enable tailored interventions for psychiatric disorders (Xiao, 2024) and improve treatment outcomes (Bansal, 2018). However, traditional linear models often fail to capture the inherent nonlinearities that define the complex dynamics of the human brain. Denoising Diffusion Probabilistic Models (DDPMs) (Ho, 2020) are a class of generative models that can produce high-quality data by learning to reverse a gradual noise-adding process through iterative denoising. In this study, we use DDPMs to model individual-specific intrinsic connectivity networks (ICNs) derived from resting-state fMRI (rs-fMRI) data.
Methods:
The data consists of rs-fMRI data from 10,000 subjects obtained from the UK Biobank database (Miller, 2016). The training process for subject-specific generation utilizes both the rs-fMRI data and their corresponding ICNs. These ICNs are generated through a fully automated, spatially constrained independent component analysis (ICA) using the NeuroMark approach (Du, 2020). The NeuroMark_fMRI_1.0 template, which includes 53 replicable ICNs derived from a 100-component blind ICA decomposition, served as a template in this ICA approach.
While training, the ICA ICNs and corresponding rs-fMRI data are input into the proposed DDPM model, as shown in Fig. 1. The DDPM operates by iteratively transforming the ICNs into a Gaussian noise distribution and then reconstructing them step-by-step through a reverse diffusion process. This helps capture the nonlinear relationships within the brain's functional connectivity and generate individualized network patterns. The model is trained over 20 epochs, with batches of 16 images. The model also employs the mean squared error (MSE) loss function, which compares the predicted noise with the actual noise added to the images. During inference, only the rs-fMRI data are used as input, allowing the model to generate the corresponding ICNs for the specific subject.
While training, the ICA ICNs and corresponding rs-fMRI data are input into the proposed DDPM model, as shown in Fig. 1. The DDPM operates by iteratively transforming the ICNs into a Gaussian noise distribution and then reconstructing them step-by-step through a reverse diffusion process. This helps capture the nonlinear relationships within the brain's functional connectivity and generate individualized network patterns. The model is trained over 20 epochs, with batches of 16 images. The model also employs the mean squared error (MSE) loss function, which compares the predicted noise with the actual noise added to the images. During inference, only the rs-fMRI data are used as input, allowing the model to generate the corresponding ICNs for the specific subject.
·Fig. 1: The proposed architecture of the DDPM model for subject-specific ICN generation.
Results:
The performance of the proposed DDPM framework is analyzed in Fig. 2(A) by evaluating the within-subject consistency for an ICN. The reconstructed networks for the same subject at two different time points exhibited high correlations, with the within-subject similarity scores averaging 0.90. During this analysis, rs-fMRI data with 490 time points were used, with the first half of the time points employed to generate one ICN and the second half used to generate the second ICN. Moreover, DDPMs effectively distinguished between individuals, as demonstrated by significantly lower similarity scores in between-subject comparisons, as shown in Fig. 2(B), with an average of 0.8 and a p-value less than 0.05. Here we observe that the within-subject stability is greater than the between-subject variability, emphasizing its ability to model the stable, individualized brain networks more effectively. Finally, multiple DDPM and ICA ICNs from different subjects were analyzed, and voxel-wise variance was calculated across subjects. Fig. 2(C) shows the heatmaps that were generated from the computed variance for both ICA and DDPM. Here ICA captures more global linear variance, reflecting shared patterns across subjects, whereas DDPM captures individualized, non-linear variability with comparatively lesser variance.
·Fig. 2: Subject-specific performance analysis of the proposed DDPM framework.
Conclusions:
This study demonstrates the efficiency of DDPMs in capturing individualized brain networks and distinguishing within-subject stability from between-subject variability. By learning nonlinear relationships in the rs-fMRI data, DDPMs provide a robust framework for precision brain mapping. The identification of individual-specific patterns underscores DDPMs' potential in advancing personalized neuroscience. Future work includes integrating multimodal data, like structural MRI, to enhance model generalizability.
Modeling and Analysis Methods:
fMRI Connectivity and Network Modeling 1
Methods Development 2
Keywords:
Computational Neuroscience
Computing
Data analysis
Design and Analysis
FUNCTIONAL MRI
Machine Learning
Other - generative modeling
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Bansal, K., Nakuci, J., & Muldoon, S. F. (2018). Personalized brain network models for assessing structure–function relationships. Current Opinion in Neurobiology, 52, 42-47.
Ho, J., Jain, A., & Abbeel, P. (2020). Denoising diffusion probabilistic models. Advances in neural information processing systems, 33, 6840-6851.
Miller, K. L., Alfaro-Almagro, F., Bangerter, N. K., Thomas, D. L., Yacoub, E., Xu, J., ... & Smith, S. M. (2016). Multimodal population brain imaging in the UK Biobank prospective epidemiological study. Nature neuroscience, 19(11), 1523-1536.
Du, Y., Fu, Z., Sui, J., Gao, S., Xing, Y., Lin, D., ... & Alzheimer's Disease Neuroimaging Initiative. (2020). NeuroMark: An automated and adaptive ICA based pipeline to identify reproducible fMRI markers of brain disorders. NeuroImage: Clinical, 28, 102375.